On July 17, 2026, Sunday Robotics unveiled ACT-2, a manipulation model that folded laundry at a 99.1% success rate across 25 homes it had never seen before. The company stresses that it collected no data and did no fine-tuning in any of those homes, and it is precisely this lack of per-home adaptation that remains one of the hardest barriers in home robotics.
Key takeaways
- 99.1% success rate (±0.3%) — 778 successful folds out of 785 attempts.
- Tested in 25 previously unseen homes, with no per-home fine-tuning.
- Median time to fold a single garment was 2 minutes 13 seconds.
- Average fold quality was 4.72 out of 5, with 73.8% of folds rated perfect.
- Demonstration data was captured with the Skill Capture Glove, and tasks were performed by the Memo robot.
What Sunday Robotics showed
ACT-2 is the second iteration of Sunday Robotics’ manipulation model and the successor to ACT-1, shown in November 2025. It drives Memo, a mobile manipulator that in the test handled garments ranging from 40 to 135 cm. During the demonstration the robot folded nine garment types in sizes from XXS to 8XL, from small shirts to large towels and hoodies.
The company’s core claim is not the success rate itself but how it was reached. ACT-2 reportedly folded laundry in 25 homes it had not been tuned for, without collecting training data in each one separately. In home robotics, the need to adapt to a specific apartment, its lighting and object placement is usually what separates a flashy demo from a working product.
The numbers behind the demo
Across 785 attempts the robot correctly folded 778 items, a 99.1% success rate with a ±0.3% margin. The median time to fold a single item was 2 minutes 13 seconds. Success alone is not the whole story — Sunday Robotics also reports an average fold-quality score of 4.72 out of 5, with 73.8% of folds judged perfect. That distinction matters, because a robot can fold a garment in a technical sense while the result is not something you would put in a drawer.
Skill Capture Glove and the Solve framework
Sunday Robotics gathers its training data in an unusual way. Instead of teleoperating?Teleoperation: Real-time remote control of a robot by a human — the operator performs the task through the machine, without autonomy. the robot, the company uses a Skill Capture Glove worn by a person who performs the task by hand while the system records the motions and maps them onto the robot’s control policy. It ties this together in a framework it calls Solve. The approach is meant to make data collection faster and cheaper than classic teleoperation, where an operator remotely drives the robot itself.
How data flows from the glove to the robot — from human motion to autonomous execution.
Laundry folding has quietly become an informal test for home robots. Tesla showed its Optimus robot folding laundry in early 2025, though in that case the robot was remotely operated by a human, as the company later acknowledged. Other humanoid developers have run similar demos, usually in controlled settings or with partial teleoperation. Sunday Robotics’ claim of working in unseen homes with no adaptation is an attempt to set itself apart from those showcases.
Why it matters
If the results hold up under independent testing, ACT-2 shifts the question in home robotics from “can the robot do it” to “can it do it in my home, without weeks of setup.” Generalizing to unseen environments with no fine-tuning is exactly the property most flashy demos lack. Capturing data with a glove rather than teleoperation could also lower the cost of scaling skills. For now these are vendor claims — the lack of independent verification and the narrow scope of a single task mean the result reads as a promising signal, not proof of market readiness.
What’s next
- Sunday Robotics says an ACT-2 beta is due in fall 2026.
- The company raised a $165M Series B (about 660 million zloty, per Humanoids Daily) to fund further development.
Sources
- Sunday Robotics — ACT-2 Preview
- Humanoids Daily — Beyond the Viral Demo: Sunday Robotics Claims 99.1% Zero-Shot Success in Laundry Folding with ACT-2
